Time to crunch Arecibo 1.24
21 May 2012 0:37:12 UTC
Topic 196341
(moderation:
Hi, 2 days ago I receive 22 WU with time estimate of 4:48 each ( those run on GPU ). Until now the first job stuck at 37.501 with 16:16:10 computing time and a little bit less than 15 hours to go ! What should I do, the dead line for the 22 job is in 12 days ?!
Time to crunch Arecibo 1.24
)
Hi!
Theoretically the jobs should run quite a bit faster on your card. You can try to free up one CPU core (for your 8 core (threads) system, you can configure (in the web interface or in the local settings in BOINC manager) that BOINC uses only up to 87% of all virtual cores for CPU tasks ==> 7 out of 8) .
Cheers
HBE
Hi, I have go down to 75% but
)
Hi, I have go down to 75% but like yesterday the process continu to use only 0.5 of a CPU. Also there is few breakpoint, the last one was at +- 20 minutes from the beginning and if I suspend, stop, restart or hibernate my computer ( it's a laptop ) I lost almost everything ( yes it's happend yesterday night ).
Do you have leave
)
Do you have leave applications in memory selected?
Yes. This is why i am
)
Yes. This is why i am disapoint.
RE: Hi, I have go down to
)
Just to make sure ...
What is your setting for "Tasks checkpoint to disk at most every xxx seconds" under "Disk and Memory usage" in the computing preferences? The default is 60 seconds. If so, there should then be a checkpoint every minute so that you will lose at most 60 seconds worth of crunching.
Cheers
HB
Yes it is at 60 second. And
)
Yes it is at 60 second.
And by the way, in the memory usage :
use 50 % when the computer running
use 75 % when the computer is not use.
Leave the process in memory when suspend.
(Sorry it's translate because I speak french)
RE: Do you have leave
)
That setting is a Red Herring, it should be called 'leave CPU applications in memory' as GPU applications are now always removed from Memory when suspending (from Boinc 6.6.37 onwards), they should resume from their last checkpoint,
Boinc 6.6.36 would sometimes get into a race condition and switch from one Seti Cuda task to another (before they had even checkpointed), once it had done it about two or three time the GPUs of those times would run out of memory,
and the Seti Cuda app would switch into CPU fallback mode, run very very slowly and often crash the PC,
Claggy
Did not know this.
)
Did not know this.
Thanks Claggy
RE: Did not know this.
)
Actually it's from Boinc 6.6.3 that GPU tasks were removed from memory when suspending:
- client: when preempting a process, remove it from memory if:
1) it uses a coprocessor
2) it has checkpointed since the client started
3) it's being preempted because of a user action (suspend job, project, or all processing) or user preference (time of day, computer in use)
From Boinc 6.6.37 GPU tasks were also removed from memory even if they hadn't checkpointed:
- client: when suspending a GPU job, always remove it from memory, even if it hasn't checkpointed. Otherwise we'll typically run another GPU job right away, and it will bomb out or revert to CPU mode because it can't allocate video RAM
Claggy
RE: Hi, I have go down to
)
The GPU app only uses about half a CPU by design (that's the part they don't have working on the GPU yet).